On Wed, Oct 16, 2019 at 10:02:09AM -0700, Kevin Fenzi wrote:
On Wed, Oct 16, 2019 at 10:50:09AM +0200, Adrian Reber wrote:
> After a successful test of the new mirrorlist server code in staging,
> this is the next step. Try to make it work in production on one of the
> proxies. As discussed on the mailing list we are trying the new setup
> with proxy14.
>
> All the conditionals have been updated to be 'staging or proxy14' and
> the mirrorlist data generation has been enhanced to also write out the
> new protobuf based format on the backend system.
>
> At the same time this moves the pickle/protobuf generation from :55 to
> :30 as this generation takes much longer than it used to be and the
> mirrorlist server restarted at :15 was never using the new data during
> the last few months.
Bummer. Is the protobuf faster than the pickle? Or are they about the
same...
The database queries are the problem. During DB query time it is running
with 40% CPU. Once it has all the data from the CPU it runs with 100%.
Reading all the trouble you currently have with koji I am sure you are
thrilled to hear that the DB is also slow for this. The generation of
pickle or protobuf makes not difference as far as I can tell (at least
compared to the DB query time).
+1 here.
Thanks.
Adrian